On Tue, 05 Apr 2016 11:39:17 -0400 (EDT), David Miller wrote: > From: Jakub Kicinski <jakub.kicin...@netronome.com> > Date: Fri, 1 Apr 2016 22:06:37 +0100 > > > When calculating the RX buffer length we need to account for > > up to 2 VLAN tags and up to 8 MPLS labels. Rounding up to 1k > > is an relic of a distant past and can be removed. While at > > it also remove trivial print statement. > > > > Signed-off-by: Jakub Kicinski <jakub.kicin...@netronome.com> > > I disagree with the MPLS aspect of this change. > > VLAN is special, in that when the hardware supports VLAN properly, the > VLAN header doesn't eat into the MTU and is sort of "transparent". > > But MPLS doesn't work that way. > > MPLS is in the main frame and takes up MTU space.
Makes sense. RFC3032 counts MPLS label stack as Frame Payload. > Therefore I see no reason to increase the buffer length by 8 * MPLS > which is just a rediculous amount of wasted space. > > I'm not applying this without at least some more explanations about > why exactly you need to account for these values in the commit message. It's just what FW guys asked me for. I'll try to find out what their reasoning was. I have a patch queued up which gets rid of unconditionally reserving 64B for firmware prepend, I guess that made me too excited to pay attention to the fact the accounting for MPLS is indeed questionable...
